RAID 0 with 3 partitions?

[Ian Roberts]

Hi

I've not done RAID before so just wanted to get clear about this...

If I use RAID 0 can I then create 3 partitions within it. Or am I stuck with
one enormous partition?

If 3 partitions are possible I'm thinking separate partitions for OS, Apps,
Data.

Planning to use two WD 74Gb SATA Raptors drives so next questions is....

Would RAID 0 with 3 partitions be faster than the two separate disks set up
as Master C (OS) and D (Apps) then Slave E (Data).

Thanks for any input.

Ian

 

[Arno Wagner]

Hi

I've not done RAID before so just wanted to get clear about this...

If I use RAID 0 can I then create 3 partitions within it. Or am I
stuck with one enormous partition?

Ordinarily RAID creates one larger "logical" disk from smaller ones.
You should be able to treat it just like a regular disk. Depending
on the controller/software used, you might get less than this.

Most/all hardwware-RAID solutions should give you this view. Linux
software RAID does.

If 3 partitions are possible I'm thinking separate partitions for
OS, Apps, Data.

Planning to use two WD 74Gb SATA Raptors drives so next questions is....

Would RAID 0 with 3 partitions be faster than the two separate disks
set up as Master C (OS) and D (Apps) then Slave E (Data).

Depends. The only good figures I have are from Linux software RAID:
About same speed on writes (maybe a bit slower). About 180%
speed of the individual disks on mostly linear reads, up to the
practical PCI-bus limit (i.e. RAID0 with 4 disks will not be 4 times
faster today than with 2-3 disks).

For shorter or non-linear reads/writes, the answer is that it depends.
If you really need high speed, you need to investigate. If on the other
hand you are find with having everything on a single Raptor,
depending on the RAID controller/software used, RAID0 should be not
much slower that a single disk and might be significantly fater
for some access patterns.

However using a cheap RAID controller or a mediocre software
solution can slow down things considerably.

Arno

 

[Bob Knowlden]

You can partition a RAID 0 array. I have 3 partitions on mine (two NTFS, one
FAT32). This was set up with the XP utilities, although I used Partiton
Magic 8 to format the FAT32 partition, as it's larger than 32 GB. (XP
supports FAT32 partitions larger than 32 GB, it just won't format them.) I
guess that hardware RAID arrays look like single drives to the disk
utilities, including the DOS-based ones I've tried (Partition Magic, Ghost),
even though the utilities may not be certified to work with RAID systems.

The rest of your questions I can't answer. I've read a few online articles
lately ( http://www.anandtech.com/storage/showdoc.aspx?i=2101) that claim
that SATA RAID 0 is largely wasted. I suspect that what holds for gamers and
users of office applications may not apply to video editing. (As I don't do
video editing, it's just speculation.)

Address scrambled. Replace nkbob with bobkn.

 

[Folkert Rienstra]

Ordinarily RAID creates one larger

"logical" disk from smaller ones.

Nonsense.
'Striping' disk drives simulates a larger 'physical' disk to the operating system.
So does 'JBOD'.
'Mirroring' does not.

'"logical" disks' are partitions.

You should be able to treat it just like a regular disk. Depending
on the controller/software used, you might get less than this.

Less than what?

Most/all hardwware-RAID solutions should give you this view. Linux
software RAID does.




Depends. The only good figures I have are from Linux software RAID:
About same speed on writes (maybe a bit slower). About 180%
speed of the individual disks on mostly linear reads, up to the
practical PCI-bus limit (i.e. RAID0 with 4 disks will not be 4 times
faster today than with 2-3 disks).

On desktop MoBos and add-on SATA controllers.
MoBos with that many channels on the MoBo Chipset or server/
professional workstation MoBos don't have that restriction.

For shorter or non-linear reads/writes, the answer is that it depends.
If you really need high speed, you need to investigate. If on the other
hand you are find with having everything on a single Raptor,
depending on the RAID controller/software used,

RAID0 should be not much slower that a single disk and
might be significantly fater for some access patterns.

So what it comes down to is that when your access pattern is mainly
random access small file the RAID0 isn't going to help.
In that case seperating OS Apps and DATA onto different physical
drives cuts down on the headmovements per drive and with SATA even
allows concurrent access at all time, provided that the OS supports it.

 

[Arno Wagner]

Nonsense.
'Striping' disk drives simulates a larger 'physical' disk to the operating system.
So does 'JBOD'.
'Mirroring' does not.

'"logical" disks' are partitions.

There is no unified terminology. On Linux this is a block device.
On Windows I have no idea.

On the other hand on BSD a "partition" is called a "slice".

The term ''logical'' can however be used in its original meaning,
namely a thing simulating to be something else while being composed
from smaller parts or being part of a larger thing. That is the
reason I used the quotes.

[...]

On desktop MoBos and add-on SATA controllers.
MoBos with that many channels on the MoBo Chipset or server/
professional workstation MoBos don't have that restriction.

Yes, and since that is the standard case my answer is for it.
Of couse, if you use, e.g., a PCI-X controller you can get more
even with add-on cards. But all these are special cases and
have to be looked at individually, since all these solutions
have their own limits. You will not get 500MB/sec reading from
an 8-way RAID5, even though the disks may be able to do that
unless you have pretty unusual hardware.

As an exapmple I get 90MB/sec sustained linear reads from an
8 way software-RAID with two Promise 150TX4 controllers in
a 66MHZ PCI bus on a "professional" dual CPU board. One nice
thing is that this rate does _not_ level off at the end
of the disks were they are slower since they are still much
faster than the bus.

There is also the problem that many mainboards do not have
all (S)ATA channels in the chipset but some of them logically
connected to the PCI bus. Of course you will get better rates
if the controllers are on some faster internal bus.

All in all the answer to the speed question is that it
depends very much on the set-up. Server mainboards are
not necessarily faster. They are usually more reliable and
more expensive.

So what it comes down to is that when your access pattern is mainly
random access small file the RAID0 isn't going to help.

Yes. Or at least it depends very much on the disk-sheduling used.
I have no figures on this.

In that case seperating OS Apps and DATA onto different physical
drives cuts down on the headmovements per drive and with SATA even
allows concurrent access at all time, provided that the OS supports it.

Yes, definitely. Although the improved reliability of RAID0 may be
more important. To me it is.

Arno

 

[Curious George]

I think others have answered your questions pretty well. I wanted to
share some caveats though. RAID as a whole has a lot of PITA details
that can trip you up. What seems right at first, often isn't worth
the trouble down the line.

Basically my take on RAID 0 is that it is only for limited
applications where the data is backed up frequently and/or left on the
array for short periods of time and only when you are absolutely
desperate for disk bandwidth beyond current technology. It's not an
ordinary kind of performance upgrade. Because of many factors, most
PC's and many workstations benefit more by doing without raid of any
kind (when you factor in setup, administration, cost, reliability,
etc).

1. All RAID arrays are statistically more likely to suffer a
failure/problem than a single disk. That likelihood increases with
every disk you add.

2. When a disk in a RAID 0 array fails or is removed all the data on
all the other disks is rendered useless and inaccessible. It’s also
can turn into a total mess when you have a single drive that is
failing/acting funny but the array is "up."

3. As RAID 0 does not create any ECC data writes are never verified
beyond the individual disk's onboard error checking abilities- so the
increased complexity & statistical likelihood of problems is not
offset in any way (like it is in raid 1, 3, etc.)

4. Raw throughput is increased and can be theoretically up to 3x
faster in your example. But not all aspects of disk performance are
multiplied across the board. In fact certain usage patterns show
little or no improvement with striping- and for one thing the
individual disk latency remains the same.

5. Adding multiple disks that work simultaneously to a PC can be very
taxing on its power supply and cooling system. Disks & PC's with
improper cooling or erratic power are less reliable.

6. If you use software/OS based striping you won't be able to boot off
the array. Many low-end RAID controllers have crappy drivers or
limited management features which can make you crazy when you just
need it to work.

7. Finally the Raptors are WD's first try at "enterprise" SATA drives.
They might turn out to be the best thing since sliced bread but they
are now somewhat unproven. The results of storagereview.com's
reliability survey (if you put any stock in it) is not overly exciting
placing that model in the lower 35th percentile (with 41 responses)
and the 36gig version in the 61st percentile (with 87 responses).

If you're still reading and want some advice I'd say stick with a
single WD or Seagate and spend your time on a good automated backup
strategy. After all there's more to backup than having a 2nd copy -
and even that is usually neglected.

If you have a specific problem or project where you know there is no
current model drive that can do the job by itself, then we should keep
talking about striping.

I suspect what you are looking for is the most responsive PC possible.
Try load-balancing approaches in your design instead. You have the
right idea of partitioning but if you separate the OS, Apps, and Data
among three ordinary drives you will improve responsivess. Fact is
even the fastest striped arrays can slow to a crawl if you throw too
many disk intensive tasks at it at the same time. Also if there is a
disk problem, it is easier to troubleshoot, isolate, and repair or
perform data recovery if necessary.

Cheers

 

[Jure Sah]

Hello,

I have here 2 WD 40 GB drives in RAID0 and an extra drive that holds the
system. The RAID0 array is split into 3 partitions (d:, f: and g,
while the extra system drive is split in 2 partitions (c: and e.

I have Windows 98 SE, so no special software buffering and the RAID0 is
on a cheap RAID controller card.

The c: and e: partitions (containing the system and some archives) are
acting slow (the more fragmetned the worse), something like 64 kbps
(partition to partition) to 128 kbps (partition to network or memory),
while the RAID0 partitions (d:, f: and g: containing all program files,
the swap file [makes software load and CDs burn FAST], all personal
files and all junk files) are very fast, if they are defragmented or not
makes no diffirence, around 96 kbps (partition to partition) to 256 kbps
(partition to network or memory).

Now while I likely converted the units wrong, the RAID0 I have on two
drives is around 4 times faster than the plain disk one.

And yes, it's quite easy to have as many partitions as you want on a
RAID array. Each RAID array acts exactly like one disk to the OS (just
faster/whatever of course).

 

[Jure Sah]

So what it comes down to is that when your access pattern is mainly
random access small file the RAID0 isn't going to help.
In that case seperating OS Apps and DATA onto different physical
drives cuts down on the headmovements per drive and with SATA even
allows concurrent access at all time, provided that the OS supports it.

Nonsense. You read bechmarks too much.

1. Moving drives from the main controller to a RAID card can boost
things like CD writing anyhow.

2. Operating systems without large disk caches can make use of the
convenient disk buffer usualy present on a RAID controller.

3. Most software is composed of blocks that load off the disk in one
chunk. With a defined RAID0 strip width of about 1 kb, and a
non-prehistorc OS, files too small to benerfit from the speed gained
with RAID0 simply do not exist.

4. Systems utilizing anything that can relate to the Windows swap file,
gain awesomely in preformance, because the swap file is usualy a 700+ MB
block. In windows, when resizing of this file occurs, it's being read
and rewritten in one go, the speed of this imporves greately with RAID0.
Even on systems where swap file resizing never occurs, synchronization
between the memory and the swap file is preformed in one go likewise and
the preformance gain is rather great.

5. Many home users can also greately benerfit from RAID0, as movies and
other digital multimedia mostly represents very big files, which can
thus be opened and copied around much faster.

6. Any system that utilizes Direct Memory Access for disk access
(practicaly ANY non-prehistoric system), does not actually preform as
many seeks, but randomly reads/writes to the memory then synchronizes
the memory with the harddrive, again in one go, utilizing a RAID0
configuration even more (only true in hardware RAID implementations).

It is greately decieving to claim that RAID0 is not actually usefull to
anyone who does not deal with large files often (heck, most home users
don't even know they're actually dealing with huge huge files when they
copy around digital movies). Sure I can understand it's recomeneded to
people who really render huge huge files onto their harddrives every 2
minutes, but... With the fact that most home users could have great
preformance, simply by plugging their harddrives elsewhere or flipping a
switch in BIOS, claiming there is no preformance boost is simply not ok.

 

[Arno Wagner]

There is no unified terminology. On Linux this is a block device.
On Windows I have no idea.

On the other hand on BSD a "partition" is called a "slice".

The term ''logical'' can however be used in its original meaning,
namely a thing simulating to be something else while being composed
from smaller parts or being part of a larger thing. That is the
reason I used the quotes.
[...]
Yes, definitely. Although the improved reliability of RAID0 may be
more important. To me it is.

Oops, sorry about that. Momentary confusion with RAID1. Actually
RAID0 is far less reliable than a single drive. If yo have drives
to spare I would advise to go to RAID1. As fast as RAID0 on reads
but a lot more reliable.

Arno

 

[Folkert Rienstra]

Nonsense.
'Striping' disk drives simulates a larger 'physical' disk to the operating system.
So does 'JBOD'.
'Mirroring' does not.

'"logical" disks' are partitions.

There is no unified terminology. On Linux this is a block device.
On Windows I have no idea.

On the other hand on BSD a "partition" is called a "slice".

The term ''logical'' can however be used in its original meaning,
namely a thing simulating to be something else while being composed
from smaller parts or being part of a larger thing. That is the
reason I used the quotes.

[...]

On desktop MoBos and add-on SATA controllers.
MoBos with that many channels on the MoBo Chipset or server/
professional workstation MoBos don't have that restriction.

Yes, and since that is the standard case my answer is for it.

Which you failed to make clear.

Of couse, if you use, e.g., a PCI-X controller you can get more
even with add-on cards. But all these are special cases and
have to be looked at individually, since all these solutions
have their own limits. You will not get 500MB/sec reading from
an 8-way RAID5, even though the disks may be able to do that
unless you have pretty unusual hardware.

Having 8 drives in RAID-5 on a desktop is what I would call unusual.

As an exapmple I get 90MB/sec sustained linear reads from an
8 way software-RAID with two Promise 150TX4 controllers in
a 66MHZ PCI bus on a "professional" dual CPU board.

Sounds awful to me. You must be doing something wrong.

One nice thing is that this
awful

rate does _not_ level off at the end of the disks were they are
slower since they are still much faster than the bus.

90MB/s is way below the capacity of that bus.

There is also the problem that many mainboards do not have
all (S)ATA channels in the chipset but some of them logically
connected to the PCI bus. Of course you will get better rates
if the controllers are on some faster internal bus.

Like Fast/Wide (66MHz/64-bit) PCI or PCI-X or PCI-Express.

All in all the answer to the speed question is that it
depends very much on the set-up.

Server mainboards are not necessarily faster.

Yes they are, although faster in server type usage may be character-
ized more like "less slow" than with running on a desktop system.

They are usually more reliable

Just another chipset from the same mfgr of MoBo chipsets.
PCI64/66 used to be an addon chip to a standard MoBo chipset.

 

[Folkert Rienstra]

It is misleading when such term is already being used by e.g. Fdisk and
-because of that- probably by any other partitioning type software.

namely a thing simulating to be something else while being composed
from smaller parts or being part of a larger thing. That is the
reason I used the quotes.
[...]
Yes, definitely. Although the improved reliability of RAID0 may be
more important. To me it is.

Oops, sorry about that. Momentary confusion with RAID1. Actually
RAID0 is far less reliable than a single drive. If yo have drives
to spare I would advise to go to RAID1.

As fast as RAID0 on reads

Only for accesses that are multiple IOs and a RAID1 imple-
mentation that spreads such multiple IO evenly over the drives.

 

[Curious George]

Right. this is called a logical disk
no. Partitions are paritions. partitions hold logical volumes.

It is misleading when such term is already being used by e.g. Fdisk and
-because of that- probably by any other partitioning type software.

You should not be so easily misled. 'logical disk' is a standard term
used by raid manufacturers. Raid manuals both use and explain this
term.

 

[Jure Sah]

The rest of your questions I can't answer. I've read a few online articles
lately ( http://www.anandtech.com/storage/showdoc.aspx?i=2101) that claim
that SATA RAID 0 is largely wasted. I suspect that what holds for gamers and
users of office applications may not apply to video editing. (As I don't do
video editing, it's just speculation.)

That article is mostly rubbish.

Try any benchmark, shows no diffirence between RAID0 and normal disks,
but try booting from it, copying movies around it or transfering over
LAN from it and suddenly RAID0 beats normal disks by a factor of 4.

I've tried a bechmark program on my computer, it decided my RAID0 array
was slower than that stand-alone disk I boot from, while it's perfectly
evident that a program loads from the stand-alone disk in 10 seconds
while from the RAID0 array in less than two (Paint Shop Pro 7).

My point: Bechmarks are bullshit, try it yourself.

The other thing perfectly true is that RAID0 can get you accross much
cheaper than stand-alone disks.

1. RAID0 is the cheapest way to get a home user 500 Gigs of disk space
in one piece.

2. You don't have to buy very expensive drives to get peak preformance
from them and the controller card is plain cheap (if RAID0 isn't already
supported by your motherboard).

3. You never ever have to defragment a RAID0 array, preformance does not
degrade with file fragmentation as with stand-alone drives and the disk
contens take much more time to fragment as much as in stand-alone drives.

4. RAID0 is an old and well-tested technology and does not cause
dissaster with CD-ROMs and uncompatible harddrives as for example the
Prefetch technology or any other weird preformance boosters available in
mobos.

 

[Arno Wagner]

Bob Knowlden wrote:

That article is mostly rubbish.

Try any benchmark, shows no diffirence between RAID0 and normal disks,
but try booting from it, copying movies around it or transfering over
LAN from it and suddenly RAID0 beats normal disks by a factor of 4.

I've tried a bechmark program on my computer, it decided my RAID0 array
was slower than that stand-alone disk I boot from, while it's perfectly
evident that a program loads from the stand-alone disk in 10 seconds
while from the RAID0 array in less than two (Paint Shop Pro 7).

That speed-up seems rather unlikely, unless the I/O sheduling of
the OS is really, really broken. Was this perhaps a situation
were the fast start was not the first from that disk and parts
still in the cache memory of the OS?

My point: Bechmarks are bullshit, try it yourself.

Benchmarks are specific. Ni general-purpose benchmarks exist or can
exist, since access patterns are vastly different for different uses
and disk-layouts (and fragmentation state).

The other thing perfectly true is that RAID0 can get you accross much
cheaper than stand-alone disks.

Huh? And why is that?

1. RAID0 is the cheapest way to get a home user 500 Gigs of disk space
in one piece.

Why do you want it in one piece? Are you a fan of fragmentation,
losses due to large clusers and catastrophic single partition
failures?

2. You don't have to buy very expensive drives to get peak preformance
from them and the controller card is plain cheap (if RAID0 isn't already
supported by your motherboard).

Or OS. Still peak performance is what you get at the price
of negative peake reliability. It is not called RAID-ZERO
for no reason.

3. You never ever have to defragment a RAID0 array, preformance does not
degrade with file fragmentation as with stand-alone drives and the disk
contens take much more time to fragment as much as in stand-alone drives.

This is nonsense. A RAID0 degrades just like other arrangements
when fragmented. It also fragments just as much as any other
arrangement of the same size, since the part of the OS that
does fragmentation does not know the difference.

4. RAID0 is an old and well-tested technology and does not cause
dissaster with CD-ROMs and uncompatible harddrives as for example the
Prefetch technology or any other weird preformance boosters available in
mobos.

Well, yes. If you are willing to live with very low reliability
and the very real possibility to loose the whole array. If it is
just unimportant stuff or ar disk-based buffer, RAID0 has some merits.
However even calling it 'storage' may be overoptimistic.

Arno

 

[Curious George]

Well, yes. If you are willing to live with very low reliability
and the very real possibility to loose the whole array. If it is
just unimportant stuff or ar disk-based buffer, RAID0 has some merits.
However even calling it 'storage' may be overoptimistic.

Arno

Indeed. I agree with your points Arno. I think this is the most
important one. MTBF of an array is equal to the MTBF of an individual
drive divided by the number of drives in the array. When 1 disk has
problems so does the entire striped set.

As far as performance I wonder if some of the performance limitations
are related to lack of spindle synchronization (rotational position
locking). Rather than explaining here see:
http://www.uni-mainz.de/~neuffer/scsi/what_is_raid.html
under "2. Data Striping" 3rd paragraph

Hard disk manufacturers argue that spindle synchronization is no
longer necessary due to faster spindle speeds and more intelligent
firmware. I wonder if it just due to the lack of popularity of levels
like 3 and the can of worms RPL tends to open up (I think there are
licensing & acceptance issues with IBM's master-independent solution).

 

[Folkert Rienstra]

I agree with your points Arno.

You like living dangerously?

I think this is the most important one. MTBF of an array is equal to the
MTBF of an individual drive divided by the number of drives in the array.

But that still says nothing about when or if your drives will die.

When 1 disk has problems so does the entire striped set.

So it doesn't make it better than a single drive, but not worse either.

As far as performance I wonder if some of the performance
limitations are related to lack of spindle synchronization
(rotational position locking). Rather than explaining here see:
http://www.uni-mainz.de/~neuffer/scsi/what_is_raid.html
under "2. Data Striping" 3rd paragraph

Hard disk manufacturers argue that spindle synchronization is no
longer necessary due to faster spindle speeds and

more intelligent firmware.

Like maybe that drives reorder reading/writing blocks within a command
so that no revs are lost when the first block position in the command was
just missed on the media and the drive just reads/writes from the position
that it is at and reads/writes the 'lost' blocks at the end of the command.

Or that the Raid firmware just handles drives on a come first basis instead
of in a fixed order.

 

[Folkert Rienstra]

Wotanidiot.

That article is mostly rubbish.

Try any benchmark, shows no diffirence between RAID0 and normal disks,
but try booting from it, copying movies around it or transfering over
LAN from it and suddenly RAID0 beats normal disks by a factor of 4.

I've tried a bechmark program on my computer, it decided my RAID0 array
was slower than that stand-alone disk I boot from, while it's perfectly
evident that a program loads from the stand-alone disk in 10 seconds
while from the RAID0 array in less than two (Paint Shop Pro 7).

My point: Bechmarks are bullshit, try it yourself.

The other thing perfectly true is that RAID0 can get you accross much
cheaper than stand-alone disks.

1. RAID0 is the cheapest way to get a home user 500 Gigs of disk space
in one piece.

2. You don't have to buy very expensive drives to get peak preformance
from them and the controller card is plain cheap (if RAID0 isn't already
supported by your motherboard).

3. You never ever have to defragment a RAID0 array, preformance does not
degrade with file fragmentation as with stand-alone drives and the disk
contens take much more time to fragment as much as in stand-alone drives.

4. RAID0 is an old and well-tested technology and does not cause
dissaster with CD-ROMs and uncompatible harddrives as for example the
Prefetch technology or any other weird preformance boosters available in
mobos.

 

[Arno Wagner]

But that still says nothing about when or if your drives will die.

So it doesn't make it better than a single drive, but not worse either.

Quite to the contrary. In RAID0 the MTBF for data loss is the
reliability of the overall array. You still have one working disk
if one of a pair breaks, but _all_ data is lost, effectively halving
the MTBF. The thing is that in RAID0 the data on one disk also breaks
if the other disk fails.

With two independent disks, you have the same probability of experiencing
data loss as result of a disk failure as in a RAID0 of two disks,
however you will only loose half the data in a one-disk failure.

If you normalize this on capacity, RAID0 with a pair of disks has
half the reliability per bit than a pair ofindependent disks.

Arno

 

[Odie Ferrous]

But that still says nothing about when or if your drives will die.


So it doesn't make it better than a single drive, but not worse either.

Come on, Folkert - how can you say a single drive is as susceptible to
failure as RAID 0 with 2 drives?

You been at the spirits again?


Odie

 

[Curious George]

But that still says nothing about when or if your drives will die.

True but that's because there isn't a nice way of
characterizing/predicting that. MTBF, with all it's flaws is a
standard statistical relibility term. How exactly would you describe
the statistical reliability of drive arrays?

So it doesn't make it better than a single drive, but not worse either.

How do you come to that?